Research map / Architectures and efficiency / RNNs and normalization
Learning and Generalization in Overparameterized Neural Networks, Going Beyond Two Layers
Zeyuan Allen-Zhu, Yuanzhi Li, Yingyu Liang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train an overparameterized two-layer ReLU network with SGD on its hidden weights | Two-layer ReLU learner · Agnostic PAC learning of two-layer smooth-activation target functions | population risk | Improved |
| Train an overparameterized three-layer ReLU network with SGD variants that allow hidden-layer interactions | Three-layer ReLU learner · Agnostic PAC learning of three-layer smooth-activation target functions | population risk | Improved |
| Use overparameterization so the learner can represent larger target functions and more complex activations | SGD-trained neural networks · Generalization to the population distribution | population risk | Improved |
| Use weight decay to scale down the three-layer network during training | Three-layer ReLU learner · Three-layer learning | optimization stability | Improved |
| Use an explicit regularizer that encourages weights to be spread across neurons | Three-layer ReLU learner · Generalization of three-layer networks | generalization | Improved |
| Train only the output layer rather than allowing hidden-layer interactions | Three-layer ReLU learner · Learning correlations between nonlinear measurements | learnable function class | Hurt |
Closest papers by meaning
- SGD Learns Over-parameterized Networks that Provably Generalize on Linearly Separable Data (ICLR 2018 · similarity 0.85)
- The Role of Over-Parameterization in Machine Learning – the Good, the Bad, the Ugly (AAAI 2024 · similarity 0.83)
- Fine-Grained Analysis of Optimization and Generalization for Overparameterized Two-Layer Neural Networks (ICML 2019 · similarity 0.83)
- Tilting the Odds at the Lottery: the Interplay of Overparameterisation and Curricula in Neural Networks (ICML 2024 · similarity 0.82)
- Do Neural Networks Need Gradient Descent to Generalize? A Theoretical Study (NeurIPS 2025 · similarity 0.82)
- How Uniform Random Weights Induce Non-uniform Bias: Typical Interpolating Neural Networks Generalize with Narrow Teachers (ICML 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
